StradVision, Inc. A San Jose-based Vision Processing Technology Company, Uses Deep-Learning to Help Bring Cutting-Edge Self-Driving Vehicle Technology to the Masses
The Silicon Review
“Our primary goal is to create a safe environment for drivers and passengers.”
StradVision is an industry pioneer in vision processing technology, accelerating the advent of fully autonomous vehicles. With more than 100 employees in its San Jose, Seoul, Munich, and Tokyo offices, StradVision addresses the industry’s challenges via its expertise in deep learning, embedded platforms, perception and advanced algorithm on a daily basis.
The company was incorporated in 2014 and is headquartered in San Jose, CA and Seoul, South Korea.
StradVision’s software is built lean, light, and flexible enough to be implemented into a variety of hardware in the shortest time possible. The importance of the size of the algorithm is often glossed over, but it is the most important element when it comes to running a deep learning-based algorithm on embedded systems.
Junhwan Kim: Interview Highlights
Rewind: Taking Flight
StradVision was set up to bring the safest automotive vision technology to as many people as possible. We don’t believe in a $5,000 Advanced Driver Assistance Systems (ADAS) option and wish to make cutting-edge technology available and accessible to everyone, and to positively affect technology adoption, and acceptance toward a safer future.
This strong direction leads our efforts to develop the most advanced embedded perception technology to enable SVNet to be available on mass production vehicles that can be realistic and accessible for everyday people. Instead of focusing on a wide array of technologies, we focus on camera-based embedded systems that are both technically and commercially the most realistic, and we will continue to be at the core of almost all ADAS & Autonomous Vehicles (AV) use cases.
Our automaker customers recognize the importance of camera-based perception. Despite new sensors and technologies, at the end of the day the discussion is regarding safety, reliability, and commercial readiness via the camera-based embedded perception. That’s why we’re laser-focused on a safe, reliable, and viable solution the market is seeking, which is SVNet.
Setting Benchmarks: Above and Beyond
Initially, StradVision’s founders chose the smart glasses and smartphone sectors to sell perception software. But despite high expectations, the smart glasses sector didn’t gain enough traction for us to grow in parallel. To make things worse, the smartphone sector also failed to deliver high enough of a demand regarding embedded perception solutions.
So, our founders had to find a solution quickly and looking at the overall technological landscape, as well as each of their market demands, and the options narrowed down to either drones or autonomous vehicles. Though it was a tough choice, it became apparent there was healthy interest from automakers. In addition, from a policy side (such as Euro NCAP), there was ample evidence from a national and regional level to push ADAS & AV technology on a national level.
Assessing the technological, commercial, and policy landscape of the ADAS & AV sector, the founders quickly realized that SVNet is a perfect fit, since it enables the market to quickly and safely implement and execute technology. From that point on, StradVision became a fully embedded perception software company for the ADAS & AV sector, focusing solely on enabling cameras to recognize objects outside the car and providing the fundamental backbone for autonomous driving. A takeaway from this experience is to always look and try to expand beyond the defined market by objectively assessing potential customers’ demand for the applied technology.
Elements: That Aid StradVision, Inc. Outshine the Market
Deep Learning IPs: We have about 200 Intellectual Properties regarding the core aspects of deep learning. This is what differentiates SVNet from other perception software, especially regarding the network size and accuracy.
Technical Maturity: Our experience installing/embedding SVNet on various hardware, as well as optimizing our algorithm onto target hardware, is second to none. This is recognized by all partners we engage with. Whether they are R&D or mass production, we deliver our solutions fully matured and optimized.
Commercial Maturity: StradVision is one of the few Deep Learning-based perception software providers with ample commercial experience and maturity. With experience accommodating our clients’ needs, as well as becoming fully compliant with various industry standards such as ASPICE and Guobiao (GB), we can deliver the industry’s highest quality Deep Learning-based software.
Trust: Most partners we cooperate with have no experience working with Deep Learning-based solutions. Whether the partnership is R&D or full production, the level of trust involving Deep Learning perception and to bring them to full commercialization, on top of all standardized systems and compliance, is crucial. The industry recognition, rapport and trust are one of the strongest assets we have.
Culture: We believe that each staff member at StradVision is an MVP in their own field. With a “just do it” mentality, technical expertise, and transparent culture, we strive to be the best in the business with the most professional and happy staff.
Staying Relevant to Consumer Interests: The Right Way
Our customers look for various new features, whether they’re driven by policy or market demand. Whether it is predicting where a pedestrian will go in the future, based on his movements, to detecting animals, we constantly inquire into SVNet’s potential application and develop/optimize the solution accordingly.
Moreover, StradVision’s primary goal is to create a safe environment for drivers and passengers. To make this a possibility, the company is solely focusing on perfecting the camera-based embedded perception algorithm critical to both ADAS & AV use cases.
Products: Ready to be Launched
Our Auto Labeling Tool (ALT) is scheduled to become commercially available very soon. Even to this day, data labeling, which is a job to teach perception algorithms to distinguish various objects, is done manually by people drawing boxes around objects and drawing lines on traffic lanes. Though the technology is cutting edge, the grunt work is done in countries like Africa, India, China, and the Philippines. StradVision’s ALT can annotate 97% of objects automatically, drastically reducing needed manpower. It can also annotate/ label objects up to 8x faster than manual work so this can exponentially scale data work for our partners.
We also have quite a few new features that will be operational soon that could potentially save many lives. Each year in the USA alone, there are more than 1 million deer-related accidents, and hundreds of deaths. Our latest product, Animal Detection, could potentially save lives by identifying animals and bringing vehicles to safety by either stopping the car or moving the vehicle to a safe spot. Also, we’re working on pose estimation that estimates people’s movement and behavior to predict where the person will be in the future — this can be applied to anticipate pedestrians on the side of the road to proactively adjust the vehicle and drastically reduce the chance of a pedestrian collision.
We already have a large volume of vehicles on the road in China and we will continue focusing on various production projects we have around the world, most notably in North America, Europe, and China. While doing so, we will continuously establish more partnerships and increase the volume of our software being used in various types of vehicles, while simultaneously and aggressively expanding in North America, Europe, and China, to sustain the growth rate.
Leadership | StradVision, Inc.
Junhwan Kim, Ph.D./CFA: Junhwan Kim serves as the Chief Executive Officer of StradVision..
Hongmo Je, MA: Hongmo Je serves as the Chief Technology Officer of StradVision.
Bongjin Jun, Ph.D.: Dr. Bongjin Jun is the Founder of StradVision and is currently an R&D Director
“With both experiences accommodating our clients’ needs, as well as becoming fully compliant with various industry standards such as ASPICE and Guobiao (GB), we can deliver the highest quality Deep Learning-based software the industry can offer.”